We propose a Bayesian approach for estimating branching tree mixture models to compare drug-resistance pathways (i.e. patterns of sequential acquisition of resistance to individual antibiotics) that are observed among Mycobacterium tuberculosis isolates collected from treatment-naïve and treatment-experienced patients. Resistant pathogens collected from treatment-naïve patients are strains for which fitness costs of resistance were not sufficient to prevent transmission, whereas those collected from treatment-experienced patients reflect both transmitted and acquired resistance, the latter of which may or may not be associated with lower transmissibility. The comparison of the resistance pathways constructed from these two groups of drug-resistant strains provides insight into which pathways preferentially lead to the development of multiple drug resistant strains that are transmissible. We apply the proposed statistical methods to data from worldwide surveillance of drug-resistant tuberculosis collected by the World Health Organization over 13 years.

MOTIVATION: Modern HIV-1, hepatitis B virus and hepatitis C virus antiviral therapies have been successful at keeping viruses suppressed for prolonged periods of time, but therapy failures attributable to the emergence of drug resistant mutations continue to be a distressing reminder that no therapy can fully eradicate these viruses from their host organisms. To better understand the emergence of drug resistance, we combined phylogenetic and statistical models of viral evolution in a 2-phase computational approach that reconstructs mutational pathways of drug resistance. RESULTS: The first phase of the algorithm involved the modeling of the evolution of the virus within the human host environment. The inclusion of longitudinal clonal sequence data was a key aspect of the model due to the progressive fashion in which multiple mutations become linked in the same genome creating drug resistant genotypes. The second phase involved the development of a Markov model to calculate the transition probabilities between the different genotypes. The proposed method was applied to data from an HIV-1 Efavirenz clinical trial study. The obtained model revealed the direction of evolution over time with greater detail than previous models. Our results show that the mutational pathways facilitate the identification of fast versus slow evolutionary pathways to drug resistance. AVAILABILITY: Source code for the algorithm is publicly available at http://biorg.cis.fiu.edu/vPhyloMM/